메뉴
Diagnose Claude Code native Remote Control safely inside the minmaxing harness. In this project, /remote-control is a readiness/troubleshooting skill; start the live native Remote Control server from a shell with claude remote-control.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | remote-control |
| description | Diagnose Claude Code native Remote Control safely inside the minmaxing harness. In this project, /remote-control is a readiness/troubleshooting skill; start the live native Remote Control server from a shell with claude remote-control. |
| argument-hint | ["name/status/troubleshoot"] |
| disable-model-invocation | true |
Use Claude Code's native Remote Control. This project route is a readiness and troubleshooting skill; in other words, a readiness and troubleshooting skill only. It does not start a live Remote Control session.
In this harness, typing /remote-control runs this skill, so it can shadow any
native slash command with the same name. To actually activate live Remote
Control, run the native CLI server command from a shell:
claude remote-control
Then open https://claude.ai/code or the Claude mobile app and connect to the
local session.
Claude Code exposes the native Remote Control server from the CLI:
claude remote-control--name, --permission-mode, --spawn, and --capacity/remote-controlThe remote surface is claude.ai/code or the Claude mobile app. The Claude Code
process keeps running locally on this machine, with the same filesystem, tools,
MCP servers, project settings, hooks, and permissions as the local session.
Do not rely on claude --remote-control or claude --rc unless your local
claude --help explicitly lists those flags. Claude Code 2.1.118 exposes
claude remote-control as the native live server command.
Do not implement a custom remote server, websocket bridge, HTTP tunnel, or MCP control plane. Do not add a browser automation backdoor for this route. Native Remote Control uses outbound HTTPS through Anthropic and does not require inbound ports from this harness.
Do not confuse this with claude --remote or Claude Code on the web. Remote
Control drives a local Claude Code process from another device; cloud sessions
run elsewhere and do not automatically inherit this local harness environment.
claude auth login
or /login.ANTHROPIC_API_KEY, Console/API-key auth, claude setup-token, and
CLAUDE_CODE_OAUTH_TOKEN are blockers for this feature.CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC,
DISABLE_TELEMETRY, CLAUDE_CODE_USE_BEDROCK, CLAUDE_CODE_USE_VERTEX,
CLAUDE_CODE_USE_FOUNDRY, or disableRemoteControl.claude in the project at least once and accept workspace trust before
expecting a remote session to attach cleanly.Run the static doctor before blaming the harness:
bash scripts/remote-control-doctor.sh --static --json
Run the no-secret contract gate before release:
bash scripts/remote-control-smoke.sh --fixtures
These checks never start a live Remote Control session and must not claim runtime proof. They only verify that committed project settings, docs, scripts, fixtures, and eval metadata do not block or misrepresent native RC.
From inside this harness session, /remote-control is diagnostic only:
/remote-control
To start live Remote Control, run this from a shell in the project root:
claude remote-control
For this trusted-local workspace, a typical explicit launch is:
claude remote-control --name ultimateminimax --permission-mode bypassPermissions
Keep that command running, then connect from https://claude.ai/code.
Use claude --version and /status locally when troubleshooting account,
organization, or auth state. If Claude reports API-key, Console, third-party
provider, inference-only token, stale organization, or disabled-policy status,
fix that auth state first.
This workspace defaults to trusted-local bypassPermissions for the operator's
solo workflow. Remote Control exposes that same local session authority from
web/mobile, so treat an active RC session like sitting at the terminal.
Keep the committed deny rules for:
Read(./.env)Read(./.env.*)Read(./.claude/settings.local.json)Read(./.claude/*.local.json)Read(./secrets/**)Never paste Remote Control URLs, QR tokens, credentials, .env contents, or
local profile secrets into artifacts, logs, PRs, or chats. Static harness evidence is compatibility evidence, not proof that a live browser or mobile session connected.
Remote Control requires a claude.ai subscription: log in with
claude auth login and choose the claude.ai path; unset ANTHROPIC_API_KEY
for the session.Remote Control requires a full-scope login token: do not use
CLAUDE_CODE_OAUTH_TOKEN or claude setup-token; refresh with full
claude.ai login.Remote Control is not yet enabled for your account: unset
CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC, DISABLE_TELEMETRY,
CLAUDE_CODE_USE_BEDROCK, CLAUDE_CODE_USE_VERTEX, and
CLAUDE_CODE_USE_FOUNDRY, then log out and back in if needed.Remote Control is disabled by your organization's policy: check /status
and the Team/Enterprise admin toggle. A managed disableRemoteControl
setting is a policy decision, not a harness bug.Remote credentials fetch failed: rerun with claude remote-control --verbose
after confirming claude.ai login, active subscription, and outbound HTTPS
access to Anthropic on port 443.Create a taste-to-artifact visualization package for a project, task, draft SPEC, UI, backend, agent runtime, dashboard, game, or product idea without implementing it. Use when the user asks to visualize taste.md, taste.vision, what the model thinks is being built, a product experience, UI mockup, architecture schematic, operator journey, or comprehension check before execution.
Run an approval-first variant of the minmaxing workflow: research, audit, plan, estimate, draft SPEC, visualize the intended product or operator experience, then stop with WAITING_FOR_VISUAL_APPROVAL before implementation. Use when the user wants to see and approve the model's understanding before code changes, or wants to continue or revise a saved visualization workflow.
Use this as the definitive workflow command for mutating work: default Opus 4.7 high planning and review with MiniMax-M2.7-highspeed Token Plan execution, plus explicit governed model profiles such as sonnetminimax for Sonnet 4.6 judgment with MiniMax execution. It must drive to a verified result, partial result, or blocked repair path; /opusminimax is the advanced engine underneath, not a competing daily command.
Run the power-user Sonnet 4.6 judgment plus MiniMax-M2.7-highspeed Token Plan executor route. Use when the operator wants to preserve Opus quota or has exhausted Opus, while keeping the same governed /opusworkflow lifecycle and MiniMax bounded execution.
Advanced engine behind /opusworkflow for provider split, packet, repair, and benchmark work, including the explicit sonnetminimax profile for Sonnet 4.6 judgment plus MiniMax-M2.7-highspeed execution. Normal build/fix/refactor work should use /opusworkflow.
Run the optional all-Opus sibling of /opusworkflow. Use when the operator explicitly wants Opus 4.7 as planner, executor, reviewer, and final judge with default high effort and optional max effort.